At one of Anaconda's customer events, an engineering team told a story that sounded like a win. They had compressed roughly a year of work into under two months by standardizing every workflow around a single frontier AI lab's stack. Velocity, delivered. The kind of number a CTO puts on a slide.
Then someone asked whether the work was secure.
Nobody could answer. Not the engineers, not the leadership above them. Everything the team had shipped ran through a black box behind an API, and the visibility that used to come with owning your own code had quietly gone missing. Anaconda opened the blog post announcing its newest acquisition with that story, and the reason is obvious: it is the sales pitch and the problem statement in the same breath.
On August 4, 2026, Anaconda acquired Enkrypt AI, an AI security and compliance company whose product tests models, agents, and Model Context Protocol servers before they reach production and then watches them at runtime. Neither announcement stated a price.
The Number Anaconda Led With Was Not About Anaconda
Buried in the announcement is a statistic that has nothing to do with the deal terms and everything to do with why the deal happened.
In the two months leading up to the announcement, Enkrypt AI scanned more than 268,000 tools across 25,000 MCP servers and reported more than 143,000 vulnerabilities, affecting 73% of those servers.
Some vocabulary, because this is the part that matters to anyone shipping agents. The Model Context Protocol is the open standard that lets an AI model call external tools: read a file, query a database, hit an internal API, move money. An MCP server is the process that exposes those tools. A "tool" in Enkrypt's count is one individual callable function. So the scan covered roughly eleven tools per server, and found roughly six problems per server.
Three quarters of the servers had at least one.
That ratio is the real story. MCP went from a published specification to default plumbing in under two years, and the deployment curve plainly outran the review curve. Enkrypt's scan says nothing about who wrote those servers or why. It says only that most of them shipped with something wrong in them.
Anaconda Is Assembling the Whole Lifecycle, One Acquisition at a Time
Enkrypt AI is the third piece Anaconda has bolted on in short order, and the shape of the strategy is now hard to miss.
| Acquisition | What It Added | Where It Sits in the Lifecycle |
|---|---|---|
| Outerbounds | Production-grade AI orchestration | Running workloads reliably at scale |
| Kilo Code | Agentic coding environment | Where builders and agents do the work |
| Enkrypt AI | Security, governance, compliance | Testing and guarding everything above |
Anaconda has spent more than a decade as the company that curates and packages the open source Python stack that data scientists actually install. That was a distribution business. What DeSanto is building now is a governed path from a notebook to a deployed agent, with an audit trail at every step, and the company is buying the missing pieces rather than growing them.
There is a competitive read here too. OpenAI bought the tooling millions of Python developers use daily earlier this year. The layer between a data scientist and their runtime has become strategically valuable, and the companies that own it are not letting go.
What Enkrypt Actually Sells
The product covers three stages, and each maps to a failure mode that has already happened to somebody.
- Pre-deployment red-teaming across more than 300 attack categories, so a model's weak spots surface in a test harness rather than in front of a customer
- Runtime guardrails that block jailbreaks and sensitive-data leakage while an agent is running, deployable inside a customer's own environment rather than requiring traffic to route through a public cloud
- Compliance automation that translates the NIST AI Risk Management Framework and the EU AI Act into enforceable controls instead of a spreadsheet somebody updates before an audit
The compliance piece has a hard deadline attached. Anaconda notes that the relevant portion of the EU AI Act went active on August 2, two days before the acquisition was announced. Enkrypt is also an OpenAI compliance integration partner, giving ChatGPT Enterprise customers audit and guardrail controls, and Anaconda says the same model-agnostic approach is planned for enterprises running Anthropic.
Enkrypt's research team has red-teamed models from Anthropic, Mistral, OpenAI, Gemini, and DeepSeek. According to Anaconda, it found exploitable attack categories in every one.
"Enterprises are running AI-native applications and services that already contain exploitable vulnerabilities and weaknesses, leaving them exposed to unknown risks they cannot easily address." — David DeSanto, CEO of Anaconda (Anaconda press release, August 4, 2026)
The Scanner Vendor Published the Scan Results on the Day It Was Acquired
The counterargument writes itself, and practitioners should hold it clearly.
The 143,000 figure is Enkrypt's own telemetry, published in a post co-bylined by the CEO of the company doing the buying and the CEO of the company being bought. Anaconda pointed to no independent reproduction of the scan. Neither company published a severity breakdown, a definition of what counts as a vulnerability, or a methodology note. A scanner that flags a missing rate limit and a scanner that flags an authentication bypass can both report a finding, and the difference between those two findings is the difference between a Jira ticket and an incident.
Vendor-reported vulnerability counts also carry a structural bias: the number is the marketing. A low count would not have appeared in the announcement.
None of that makes the underlying claim wrong. The direction is corroborated by events LDS has covered independently, including an AI agent that breached Hugging Face and the billion-dollar consolidation around agent identity management. Something real is being priced. The precise magnitude is being asserted by a party with an interest in the answer.
Enkrypt AI will keep operating with no changes to its existing products, plans, or customer support while the companies work toward a unified platform.
The Bottom Line
Every MCP server you stood up to unblock a demo is now a piece of production infrastructure with a callable surface. Nobody is going to schedule a review for it. The market just told you what that gap is worth by buying the company that measures it.
The gap Anaconda is selling into is not really a security gap. It is an accountability gap. A CTO can say how many tokens the org burned last month. Very few can say what their agents were permitted to touch while burning them, and fewer still can produce a document that survives contact with an auditor.
Sahil Agarwal, co-founder and CEO of Enkrypt AI, put the pitch in one line that will now be repeated at every enterprise AI procurement meeting for the next year.
"There's a difference between running AI, and running AI safely." — Sahil Agarwal, co-founder and CEO of Enkrypt AI (Anaconda press release, August 4, 2026)
The harder question is whether a platform can sell you the tooling, the runtime, the coding agent, and the thing that grades all three. Anaconda is now the vendor, the orchestrator, and the auditor. That is a lot of the stack to own before anyone outside the company has checked the 143,000.
Sources
- What It Means To Secure AI on Your Own Terms: Anaconda Acquires Enkrypt AI (Aug 4, 2026)
- Anaconda Acquires Enkrypt AI (Press Release) (Aug 4, 2026)
- Anaconda Acquires Enkrypt AI to Secure the Trillion-Token Enterprise (Aug 4, 2026)
- AI News Briefs Bulletin Board for August 2026 (Aug 4, 2026)
- Anaconda Acquires Enkrypt AI to Expand AI Security and Governance Platform (Aug 5, 2026)
- Anaconda Acquires Kilo Code to Power the Trillion-Token Enterprise (2026)
- Anaconda Acquires Outerbounds to Unify AI-Native Development (2026)